DisKit : Discovering Knowledge from Interaction Traces

DisKit implements a knowledge discovery process from traces.

For the mining step, DisKit uses dmt4sp to find frequent serial episodes from one or several sequences of events.

dmt4sp supports numerous types of constraints such as support, temporal constraints (window, gap), syntactic constraints (length, prefix, suffix), etc.
(see web page of dmt4sp)

DisKit adds many other features to better focus the exploration of traces.

Before mining, it is possible to split a trace depending on a context expressed as a set of attributes that may result in a significant reduction of combinatorial redundancy. The resulting episodes are meaningful regarding their apparition context, such as for example the actor, the object and the temporal phase of a sequence of actions.

DisKit adds other types of constraints such as the search of closed serial episodes, constraints on the presence or absence of specific "patterns" (subsequences), in order to focus on a given episode and reduce the number of resulting episodes.

DisKit outputs several interestingness measures such as event or temporal coverage, noise, and a kind of frequency to take into account overlapping episode occurrences.

Diskit can be used to process traces provided as datasets such as csv files, or traces stored in a trace-based system such as kTBS (kernel Trace-Based System), like in the Transmute prototype (see a demonstration of Transmute).